LONG-RANGE EARTHQUAKE FORECASTING BASED ON A SINGLE PREDICTOR WITH CLUSTERING

The methodology for formulating and testing a predictive hypothesis of earthquake occurrence based on a single predictor (Rhoades & Evison 1979) has been extended to allow for a cluster of similar precursors being related to a subsequent cluster of mainshock events. The hazard is conveniently represented by conditional intensity, from which probability is readily derived. and which can be updated as new information becomes available. This method has been applied to the generalized swarm hypothesis, tests of which are in progress in New Zealand and Japan. Probability distributions for the mainshock origins are estimated from magnitude-weighted locations and origin times of all the swarm earthquakes in a cluster, and from the three largest magnitudes. Hazard refinement relative to the Poisson estimate is obtained by combining the contributions for principal and secondary mainshocks. As time passes, the hazard is adjusted for further swarm occurrences, and subsequently for the occurrence or non-occurrence of mainshock events. The application of these methods is illustrated by an example from Japan, using combined regression data.